Honcho MCP Server

A remote MCP server for Honcho, providing AI memory and personalization tools to LLM clients like Claude Desktop.

Clients connect over HTTP SSE — no local process spawning, no stdio bridge needed.

Deploy

Docker (recommended)

docker build -t honcho-mcp .
docker run -d -p 3000:3000 \
  -e HONCHO_API_URL=http://your-honcho-backend:8000 \
  -e HONCHO_API_KEY=your-key \
  honcho-mcp

Manual

bun install
HONCHO_API_URL=http://your-honcho-backend:8000 HONCHO_API_KEY=your-key bun start

Related MCP server: GroundMemory

Connect

Add the server URL to your MCP client config:

Note: User identity (peer_id) is set through Honcho tool calls — the X-Honcho-* headers in the MCP config below are optional metadata and not consumed by the Honcho SDK. The server handles both SSE (GET) and JSON-RPC (POST) at the root / path.

VS Code — add to .vscode/mcp.json:

{
  "servers": {
    "honcho": {
      "type": "http",
      "url": "http://your-host:3000",
      "headers": {
        "X-Honcho-User-Name": "your-user-name",
        "X-Honcho-Workspace-ID": "default",
        "X-Honcho-Assistant-Name": "Assistant"
      }
    }
  }
}

Claude Desktop — edit claude_desktop_config.json:

{
  "mcpServers": {
    "honcho": {
      "type": "http",
      "url": "http://your-host:3000",
      "headers": {
        "X-Honcho-User-Name": "your-user-name",
        "X-Honcho-Workspace-ID": "default",
        "X-Honcho-Assistant-Name": "Assistant"
      }
    }
  }
}

Configuration

Server-side environment variables:

Client MCP config headers (use in .vscode/mcp.json or Claude Desktop config):

Available Tools

Workspace: inspect_workspace, list_workspaces, search, get_metadata, set_metadata

Peers: create_peer, list_peers, chat, get_peer_card, set_peer_card, get_peer_context, get_representation

Sessions: create_session, list_sessions, delete_session, clone_session, add_peers_to_session, remove_peers_from_session, get_session_peers, inspect_session, add_messages_to_session, get_session_messages, get_session_message, get_session_context

Conclusions: list_conclusions, query_conclusions, create_conclusions, delete_conclusion

System: schedule_dream, get_queue_status

Development

bun install
bun start        # http://localhost:3000
bun run build    # type-check